Most electric vehicle (EV) batteries on the market today contain roughly the same key materials that dictate their power, charge time, cycle life, flammability and general safety performance. In the search for lower cost and higher performance, considerable development focus has centred on the electrolyte in particular. Within this space, almost all EV battery electrolytes draw on different combinations of the same 11 molecules, a tiny fraction of the potentially 50 billion molecules available. California start-up Aionics believes that tapping artificial intelligence (AI) to evaluate a much larger pool of potential molecule combinations could revolutionise battery development and the wider EV market.
Based in Palo Alto, Aionics was founded in 2020 by three Stanford PhDs working at the intersection of AI and materials science. “We knew when we founded this company that AI algorithms would only get better over time, that data would only become more plentiful, and that compute would grow cheaper and more powerful,” says Chief Executive Austin Sendek. “That will completely change how batteries are developed for EVs.”
The company has built an end-to-end battery-design platform to engineer cells in a way to differentiate their performance, with minimal additional overhead or manufacturing cost. “Internal combustion engines (ICE) are certainly not uniform across the industry, and yet the automotive sector is more or less looking at the prospect of using roughly the same batteries,” he tells Automotive World.
The idea is that different vehicle models or markets can offer more tailored battery characteristics: luxury cars may promise longer ranges or more powerful performance, while cold-weather regions of the world focus on batteries that function well at low temperatures. The best way to do that, according to Sendek, is to engineer the electrolyte. “Essentially we’re engineering cells via the electrolyte with high-performance compute and AI,” he summarises.
Why bring in AI?
Without AI, that’s an almost impossible challenge. “There are about 50 billion molecules that a developer could be confident of procuring from a commercial vendor within a reasonable amount of time,” he explains. When combinations of molecules are considered, the possibilities are closer to ten to the 50th power. The challenge is then predicting how each of those combinations will perform within the battery.
There are numerous parameters to consider, such as the impact on cycle life, freezing point or viscosity, to name a few. A human scientists would have to evaluate that vast universe of combinations for each different parameter. This is where AI comes to the rescue.
Sendek points to a 2018 study the team published out of Stanford, which showed how AI algorithms outperform human intuition by many times over. “We took a really simple machine learning model for predicting electrolyte properties, and held it up against a team of PhD students. The model outperformed them, both in terms of speed and accuracy. That was six years ago, but it served as an early sign that there was something really powerful here,” he says.
Custom engineering
While the number of EVs on the road remains relatively tiny compared to ICE models, technology developments over the past couple of decades have been notable. In particular, there have been huge advances in battery energy efficiency and cost. So why is the industry still using only a fraction of the chemical spectrum?
“We found things that worked okay, and since then there hasn’t been a big push to broaden the scope,” suggests Sendek. “Within the research community, perhaps about 1,000 molecules have been explored for EV battery application in various publications, but even that is a tiny fraction of the whole chemical space, and we have the entire universe to explore. It’s difficult to even broach without AI.”
We have that universe of 50 billion molecules in our back pocket. We are able to assess any of those for any given application
That could change quickly thanks to the Aionics platform. In October 2023, the company announced a joint development partnership with Cellforce Group, Porsche’s wholly owned subsidiary focused on development of high-performance lithium-ion cells for automotive applications. The partnership aims to create a battery that is distinctly Porsche. While this is the only publicly announced automotive project of its kind, Sendek implies that other partnerships are also at work.
“The long-term view is that not only will we be able to achieve better performance, but we will be able to custom engineer cells for the specific markets that automotive companies are targeting,” he says. “The industry will move away from this one-size-fits-all approach.”
Off to the races
Aionics aims to map the wider chemical space and open it up for mobility. “We constantly run calculations. Our CPUs and our GPUs are always turning and generating data, learning about both the more common and the more exotic molecules, making connections between them and building up intelligence about the chemical universe,” he explains. “There’s a positive feedback loop where the more data you generate, the more you learn, and the more you learn, the better data you can generate, etc. Our goal is to know everything there is to know about this world, and so the process to design a better battery becomes asymptotically fast, cheap, and easy.”
As of now, there is still much to learn. “The chemical space is so unexplored that in some sense, we don’t know what we don’t know,” he says. Promisingly, this particular AI application could benefit from progress seen in other forms of AI, such as ChatGPT and large language models. “We’ve entered a new era of language-inspired AI, and that is percolating into our world of chemical AI,” he says.
Sendek’s Co-founder, Professor Venkat Viswanathan at the University of Michigan, has been developing some interesting chemical prediction models that use this language-based concept. “What’s going on now will make what we were doing just a few years ago look silly by comparison,” adds Sendek. “Things are changing so quickly, and it’s really been enabled by the large amounts of chemical data that are out there now.”
Today, Aionics’ focus is on improving the accuracy and speed of its AI platform. “We have that universe of 50 billion molecules in our back pocket. We are able to assess any of those for any given application,” he emphasises. “Over time we hope to make that faster and more accurate, but at least we’ve opened the door to that world, and now it’s really off to the races.”
